In [1]:
import pandas as pd
import numpy as np
from matplotlib import pyplot as plt
import seaborn as sns
import copy
In [2]:
filepath = '/Users/dangoldberg/Desktop/DataScienceInterviewProjects/Toronto Blue Jays/catcherframe.csv'
In [3]:
df = pd.read_csv(filepath)
In [5]:
df.info()
In [6]:
df.sample(20)
Out[6]:
In [15]:
called_pitches = df[(df['swings'] == 0) & (df['intentball'] == 0) & (df['hbp'] == 0) & (df['pitchout'] == 0)]
In [16]:
sns.jointplot("plate_x","plate_z",data=called_pitches,xlim=[-9,9],ylim=[-4,14],alpha=0.01)
plt.title('All Pitches')
plt.show()
sns.jointplot("plate_x","plate_z",data=called_pitches[(called_pitches['batside'] == 'R') & (called_pitches['calledstrike'] == 1)],xlim=[-9,9],ylim=[-4,14],alpha=0.01)
plt.title('RHH')
plt.show()
sns.jointplot("plate_x","plate_z",data=called_pitches[(called_pitches['batside'] == 'L') & (called_pitches['calledstrike'] == 1)],xlim=[-9,9],ylim=[-4,14],alpha=0.01)
plt.title('LHH')
plt.show()
In [19]:
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.cm as cm
import matplotlib.colors as colors
from matplotlib.lines import Line2D
from scipy.ndimage.filters import gaussian_filter
In [20]:
df['batside'].unique()
Out[20]:
In [21]:
def make_heatmap(x, y, bins):
heatmap, xedges, yedges = np.histogram2d(x, y, bins=bins, range=[[-4, 4], [0, 8]], density=False)
return heatmap, xedges, yedges
def prepare_dataset(called_pitches, bins=100, s=1):
called_balls = called_pitches[called_pitches['ball'] == 1][['plate_x','plate_z','calledstrike']]
called_strikes = called_pitches[called_pitches['calledstrike'] == 1][['plate_x','plate_z','calledstrike']]
balls_map, balls_xedges, balls_yedges = make_heatmap(called_balls['plate_x'].values, called_balls['plate_z'].values, bins=bins)
balls_map = np.flip(balls_map.T,axis=0)
strikes_map, _, _ = make_heatmap(called_strikes['plate_x'].values, called_strikes['plate_z'].values, bins=bins)
strikes_map = np.flip(strikes_map.T,axis=0)
totals_map = balls_map + strikes_map
totals_normed_map = totals_map / called_pitches.shape[0]
strike_probabilites_map = np.nan_to_num(strikes_map / totals_map)
balls_probabilites_map = np.nan_to_num(balls_map / totals_map)
probabilities_difference_map = (strike_probabilites_map * 2) - 1
strikes_map = gaussian_filter(strikes_map, sigma=s)
balls_map = gaussian_filter(balls_map, sigma=s)
totals_normed_map = gaussian_filter(totals_map, sigma=s)
strike_probabilites_map = gaussian_filter(strike_probabilites_map, sigma=s)
balls_probabilites_map = gaussian_filter(balls_probabilites_map, sigma=s)
probabilities_difference_map = gaussian_filter(probabilities_difference_map, sigma=s)
#strike_probabilites_map[strikes_map < 2] = np.nan
#balls_probabilites_map[balls_map < 2] = np.nan
#probabilities_difference_map[totals_map < 20] = np.nan
#totals_map[totals_map < 20] = np.nan
#strikes_map[strikes_map < 2] = np.nan
#balls_map[balls_map < 2] = np.nan
return strikes_map, balls_map, totals_normed_map, strike_probabilites_map, balls_probabilites_map, probabilities_difference_map, totals_map
def make_zone_subplot(values_map, fig, ax, cmap='RdBu', extent=[-4, 4, 0, 8], contour_X=None, contour_Y=None, alpha=0.7, vmin=None, vmax=None, norm=None, cbar_label='', legend=False):
if contour_X is not None and contour_Y is not None:
imc = ax.contour(X,Y,values_map, extent=extent, cmap=cmap)
#fig.colorbar(im1c, ax=ax1)
im = ax.imshow(values_map, extent=extent, cmap=cmap, alpha=alpha, vmin=vmin, vmax=vmax, norm=norm)
cbar = fig.colorbar(im, ax=ax)
cbar.set_label(cbar_label)
ax = make_zone_lines(ax)
#ax.grid(color='black', linestyle='--', linewidth=0.1, alpha=0.5)
return ax, fig
def make_zone_lines(ax, color='black', legend=False):
y_array = [1.6,3.4]
plate_left = Line2D([-2./3,-2./3],y_array,linestyle='-',color=color)
plate_left_plus = Line2D([-2./3-1.5/12,-2./3-1.5/12],y_array,linestyle='--',color=color)
plate_right = Line2D([2./3,2./3],y_array,linestyle='-',color=color)
plate_right_plus = Line2D([2./3+1.5/12,2./3+1.5/12],y_array,linestyle='--',color=color)
line1 = ax.add_line(copy.copy(plate_left))
line2 = ax.add_line(copy.copy(plate_right))
line3 = ax.add_line(copy.copy(plate_left_plus))
line4 = ax.add_line(copy.copy(plate_right_plus))
if legend is True:
ax = ax.legend((line1, line2,), ('Plate Edge','1/2 Ball Width Past Plate Edge',))
return ax
In [22]:
for side in ['RHH','LHH']:
for pit in ['RHP','LHP']:
if side == 'RHH':
batside = ['R']
elif side == 'LHH':
batside = ['L']
if pit == 'RHP':
pitcherhand = ['R']
elif pit == 'LHP':
pitcherhand = ['L']
all_pitches = df[(df['batside'].isin(batside)) & (df['pitcherhand'].isin(pitcherhand))]
bins=100
alpha = 0.7
fig = plt.figure()
ax31 = fig.add_axes([0., 0., 1.4, 1.4, ])
ax31.set_title('{} vs {} Called Balls - Observed Volume'.format(side, pit), fontdict={'fontsize': 20})
ax32 = fig.add_axes([1.5, 0., 1.4, 1.4, ])
ax32.set_title('{} vs {} Called Balls - Probability'.format(side, pit), fontdict={'fontsize': 20})
ax21 = fig.add_axes([0., 1.6, 1.4, 1.4, ])
ax21.set_title('{} vs {} Called Strikes - Observed Volume'.format(side, pit), fontdict={'fontsize': 20})
ax22 = fig.add_axes([1.5, 1.6, 1.4, 1.4, ])
ax22.set_title('{} vs {} Called Strikes - Probability'.format(side, pit), fontdict={'fontsize': 20})
ax11 = fig.add_axes([0., 3.2, 1.4, 1.4, ])
ax11.set_title('{} vs {} Called Pitches - Observed Volume'.format(side, pit), fontdict={'fontsize': 20})
ax12 = fig.add_axes([1.5, 3.2, 1.4, 1.4, ])
ax12.set_title('{} vs {} Called Pitches - Decision Certainty'.format(side, pit), fontdict={'fontsize': 20})
x = np.linspace(-4.,4.,bins)
y = np.linspace(8,0,bins)
X, Y = np.meshgrid(x,y)
strikes_map, balls_map, totals_map, strike_probabilites_map, balls_probabilites_map, probabilities_difference_map, totals_map_all_pitches = prepare_dataset(
all_pitches)
ax11 = make_zone_subplot(totals_map, fig, ax11, cmap='viridis', contour_X=X, contour_Y=Y)
ax12 = make_zone_subplot(probabilities_difference_map, fig, ax12, cmap='RdBu_r', contour_X=X, contour_Y=Y)
ax21 = make_zone_subplot(strikes_map, fig, ax21, cmap='inferno', contour_X=X, contour_Y=Y)
ax22 = make_zone_subplot(strike_probabilites_map, fig, ax22, cmap='inferno', contour_X=X, contour_Y=Y)
ax31 = make_zone_subplot(balls_map, fig, ax31, cmap='inferno', contour_X=X, contour_Y=Y)
ax32 = make_zone_subplot(balls_probabilites_map, fig, ax32, cmap='inferno', contour_X=X, contour_Y=Y)
fig.show()
Look at difference between LHH and RHH¶
In [23]:
fig = plt.figure()
#ax1 = fig.add_axes([0., 0., 1.4, 1.4, ])
#ax1.set_title('LHH Strike Probabilities', fontdict={'fontsize': 20})
ax2 = fig.add_axes([1.3, 0., 1.4, 1.4, ])
ax2.set_title('Strike Probability Delta (Pitcher View)', fontdict={'fontsize': 20})
#ax3 = fig.add_axes([2.6, 0., 1.4, 1.4, ])
#ax3.set_title('RHH Strike Probabilities', fontdict={'fontsize': 20})
alpha = 0.6
bins=120
s=2.8
x = np.linspace(-4.,4.,bins)
y = np.linspace(8,0,bins)
X, Y = np.meshgrid(x,y)
_, _, _, strike_probabilites_map_r, _, _, _ = prepare_dataset(called_pitches[(called_pitches['batside'] == 'R')], bins=bins, s=s)
#ax3, fig = make_zone_subplot(strike_probabilites_map_r, fig, ax3, cmap='Greens', contour_X=X, contour_Y=Y, vmin=0., vmax=1.0, alpha=alpha)
_, _, _, strike_probabilites_map_l, _, _, totals_map_all_pitches = prepare_dataset(called_pitches[called_pitches['batside'] == 'L'], bins=bins, s=s)
#ax1, fig = make_zone_subplot(strike_probabilites_map_l, fig, ax1, cmap='Purples', contour_X=X, contour_Y=Y, vmin=0., vmax=1.0, alpha=alpha)
strike_probabilities_delta = (strike_probabilites_map_r - strike_probabilites_map_l)
strike_probabilities_delta[totals_map_all_pitches < 1] = np.nan
strike_probabilities_delta[np.abs(strike_probabilities_delta) < 0.005] = np.nan
norm = cm.colors.Normalize(vmax=abs(strike_probabilities_delta).max(), vmin=-abs(strike_probabilities_delta).max())
ax2, fig = make_zone_subplot(strike_probabilities_delta, fig, ax2, cmap='PRGn', contour_X=X, contour_Y=Y, norm=norm, alpha=alpha, cbar_label='z = p( cs | RHH ) - p( cs | LHH )', legend=True)
text_args = {'fontweight':'ultralight', 'fontsize':12, 'horizontalalignment':'center', 'fontstyle':'italic'}
arrowprops = {'color':'grey','alpha':0.4}
ax2.annotate('More\nStrikes\nFor\nRHH', (-1.5,3.0), xytext=(-3.1,5.2), arrowprops=arrowprops, color='green', **text_args)
ax2.annotate('More\nStrikes\nFor\nLHH', (1.5,3.0), xytext=(3.1,5.2), arrowprops=arrowprops, color='purple', **text_args)
text2_args = {'fontweight':'ultralight', 'fontsize':8}
arrowprops = {'color':'grey','alpha':0.4, 'linewidth':0.001}
ax2.annotate('Rulebook Home Plate Edge', (2./3,1.6), xytext=(-1.,0.5), arrowprops=arrowprops, color='black', **text2_args, horizontalalignment='right')
ax2.annotate('½ Ball Width Outside\nof Rulebook Home\nPlate Edge', (2./3+1.5/12,1.6), xytext=(2.1,0.3), arrowprops=arrowprops, color='black', **text2_args, horizontalalignment='left')
#plt.savefig('strike_zone_delta.svg')
plt.show()
Excusion into Fastballs vs. Breaking Balls/Offspead¶
In [24]:
def is_strike_or_out(df):
if (df['calledstrike'] == 1) or (df['swings'] == 1 and df['inplay'] == 0) or (df['swings'] == 1 and df['pitcheventtype'] == 'hit_into_play_out') is True:
return 1
else:
return 0
In [25]:
df['strike_or_out'] = df.apply(is_strike_or_out, axis=1)
In [26]:
def prepare_dataset_outcomes(swinging_pitches, bins=100, s=1):
fastballs = swinging_pitches[swinging_pitches['pitchtype'].isin(['FA','FC','FS','SI'])][['plate_x','plate_z','strike_or_out']]
breaking_balls = swinging_pitches[swinging_pitches['pitchtype'].isin(['CH','CU','SL','SC','KN'])][['plate_x','plate_z','strike_or_out']]
fastballs_map, fastballs_xedges, fastballs_yedges = make_heatmap(fastballs['plate_x'].values, fastballs['plate_z'].values, bins=bins)
fastballs_map = np.flip(fastballs_map.T,axis=0)
breaking_balls_map, _, _ = make_heatmap(breaking_balls['plate_x'].values, breaking_balls['plate_z'].values, bins=bins)
breaking_balls_map = np.flip(breaking_balls_map.T,axis=0)
totals_map = fastballs_map + breaking_balls_map
totals_normed_map = totals_map / swinging_pitches.shape[0]
fastballs_probabilites_map = np.nan_to_num(fastballs_map / totals_map)
breaking_balls_probabilities_map = np.nan_to_num(breaking_balls_map / totals_map)
probabilities_difference_map = fastballs_probabilites_map - breaking_balls_probabilities_map
fastballs_map = gaussian_filter(fastballs_map, sigma=s)
breaking_balls_map = gaussian_filter(breaking_balls_map, sigma=s)
totals_normed_map = gaussian_filter(totals_map, sigma=s)
fastballs_probabilites_map = gaussian_filter(fastballs_probabilites_map, sigma=s)
breaking_balls_probabilities_map = gaussian_filter(breaking_balls_probabilities_map, sigma=s)
probabilities_difference_map = gaussian_filter(probabilities_difference_map, sigma=s)
#strike_probabilites_map[strikes_map < 2] = np.nan
#balls_probabilites_map[balls_map < 2] = np.nan
#probabilities_difference_map[totals_map < 20] = np.nan
#totals_map[totals_map < 20] = np.nan
#strikes_map[strikes_map < 2] = np.nan
#balls_map[balls_map < 2] = np.nan
return fastballs_map, breaking_balls_map, totals_normed_map, fastballs_probabilites_map, breaking_balls_probabilities_map, probabilities_difference_map, totals_map
In [27]:
for side in ['RHH','LHH']:
for pit in ['RHP','LHP']:
if side == 'RHH':
batside = ['R']
elif side == 'LHH':
batside = ['L']
if pit == 'RHP':
pitcherhand = ['R']
elif pit == 'LHP':
pitcherhand = ['L']
outcomes = df[(df['batside'].isin(batside)) & (df['pitcherhand'].isin(pitcherhand))]
bins=100
alpha = 0.7
s=1
fig = plt.figure()
ax31 = fig.add_axes([0., 0., 1.4, 1.4, ])
ax31.set_title('{} vs {} Breaking Balls - Observed Volume'.format(side, pit), fontdict={'fontsize': 20})
ax32 = fig.add_axes([1.5, 0., 1.4, 1.4, ])
ax32.set_title('{} vs {} Breaking Balls - Probability of Strike or Out'.format(side, pit), fontdict={'fontsize': 20})
ax21 = fig.add_axes([0., 1.6, 1.4, 1.4, ])
ax21.set_title('{} vs {} Fastballs - Observed Volume'.format(side, pit), fontdict={'fontsize': 20})
ax22 = fig.add_axes([1.5, 1.6, 1.4, 1.4, ])
ax22.set_title('{} vs {} Fastballs - Probability of Strike or Out'.format(side, pit), fontdict={'fontsize': 20})
ax11 = fig.add_axes([0., 3.2, 1.4, 1.4, ])
ax11.set_title('{} vs {} All Pitches - Observed Volume'.format(side, pit), fontdict={'fontsize': 20})
ax12 = fig.add_axes([1.5, 3.2, 1.4, 1.4, ])
ax12.set_title('{} vs {} All Pitches - Incremental Likelihood of Strike or Out, FB vs. BB'.format(side, pit), fontdict={'fontsize': 20})
x = np.linspace(-4.,4.,bins)
y = np.linspace(8,0,bins)
X, Y = np.meshgrid(x,y)
fastballs_map, breaking_balls_map, totals_map, fastballs_probabilites_map, breaking_balls_probabilities_map, probabilities_difference_map, totals_map_all_pitches = prepare_dataset_outcomes(
outcomes, s=s)
ax11 = make_zone_subplot(totals_map, fig, ax11, cmap='inferno', contour_X=X, contour_Y=Y)
ax12 = make_zone_subplot(probabilities_difference_map, fig, ax12, cmap='RdBu_r', contour_X=X, contour_Y=Y, vmax=1.0, vmin=-1.0)
ax21 = make_zone_subplot(fastballs_map, fig, ax21, cmap='inferno', contour_X=X, contour_Y=Y)
ax22 = make_zone_subplot(fastballs_probabilites_map, fig, ax22, cmap='RdBu_r', contour_X=X, contour_Y=Y, vmax=1.0, vmin=0.0)
ax31 = make_zone_subplot(breaking_balls_map, fig, ax31, cmap='inferno', contour_X=X, contour_Y=Y)
ax32 = make_zone_subplot(breaking_balls_probabilities_map, fig, ax32, cmap='RdBu_r', contour_X=X, contour_Y=Y, vmax=1.0, vmin=0.0)
fig.show()
